Updates

Model and report changes

  1. The model now accounts for the ongoing immunisation programme, stratifying the population of people still susceptible to infection with the virus according to their immunisation status (unimmunised/1 dose/2 doses). We use data on the daily proportions of the population getting immunised to inform this splitting of the population, assuming that it takes three weeks for vaccine-derived immunity to develop. Vaccine efficacy is assumed against both infection and death, using values for the efficacy in agreement with those found here. We have a changepoint in the vaccine efficacy on the 10th May, which marks a transition from alpha being the dominant variant, to delta.
  2. The model now also accounts for a different susceptibility to infection in each adult age group (no prior information is used); and for the under-15s, (using prior information from Viner et al, 2020, which estimates children to be less likely to acquire infection when in contact with an infectious individual).
  3. The model has the ability to incorporate estimates of community prevalence, by region and age group, from the Office of National Statistics COVID-19 Infection Survey (see Data Sources for details). These are included weekly since the outset of the Survey in May 2020 for the age groups >4 years to inform trends in incidence that are too recent to be captured by the data on deaths.
  4. The geographical definition has been changed from the seven NHS regions (map) to the nine regions typically used in government (map). This new spatial definition more appropriately reflects the existing regional heterogeneity.
  5. The underlying probability of an unvaccinated individual dying following infection with SARS-CoV2 (the infection-fatality rate, IFR) is allowed to change gradually over the course of 30 days every (approximately) 100 days. This is designed to reflect fluctuations due to seasonal effects, demand on healthcare services or the emergence of new virus variants of differing severity.
  6. The ‘Epidemic summary’ now only reports the current value for the IFR by age. To visualise how this has changed over time in our model, see the IFR tab in the ‘Infections and Deaths’ section of the report. The quantity that is now plotted under this tab is the probability of dying if infected, taking into account the impact of the immunisation programme.

Updated findings

  1. The estimate of the daily number of new infections on the 19th June across England is 15,200 (11,600–19,000, 95% credible interval). Though incidence is still increasing, this represents a slight downward revision of our most recent published estimate of 11th June.
  2. The daily infection rate is estimated to be the highest in the North West (NW) and the North East (NE) with 42 infections per 100K population per day. These correspond to 3,040 and 1,100 new daily infections respectively. The South West (SW) is the third highest with 32 infections per 100k (1,790 daily infections) followed by the West Midlands (WM) (26 per 100,000 and 1,510 new infections every day). All other regions now have more than 10 new infections per 100k population, with incidence in the South East (SE) still the lowest (18 per 100k). Note that a substantial proportion of these daily infections will be asymptomatic.
  3. We predict that the number of deaths occurring daily is likely to remain low but also likely to start increasing. For the 10th July we forecast between 32 and 79 daily deaths, though there is a distinct lack of fit to these data over the past two months with consequent very low confidence in these projections.
  4. The probability of Rt exceeding 1 is 95% in the NW, 75% in the SW and 68% in the SE. It is greater than 50% in each of NE, WM and in London (GL). This probability is lowest for the East Midlands, where it is only 31%.
  5. The growth rate for England has decreased to 0.01 (0.00–0.02, 95% credible interval) per day. This means that, nationally, the number of infections is highly likely to be increasing, although there is considerable uncertainty and heterogeneity across regions, with possible negative growth in some regions. This rate of growth corresponds to a doubling in the number of new infections every 78.9 days
  6. London, followed by the NE and the West Midlands (WM), has the highest attack rates, that is the proportions of the regional populations who have ever been infected, with 32%, 28% and 26% respectively. The SW continues to have the lowest attack rate at 15%. These attack rates are entirely consistent with our previous published report.
  7. Note that the deaths data used are only very weakly informative on Rt over the last two weeks and are thankfully sparse. Therefore, the estimate for current incidence, Rt and the forecast of daily numbers of deaths are likely to be subject to some revision.

Interpretation

The plots of the estimated Rt over the most recent weeks show that the Rt appear to have peaked three weeks ago and have declined slightly since then, in the absence of any further major relaxation of pandemic mitigation measures. The NW had the highest peak in Rt of 1.36 (CrI 1.15–1.57) with a subsequent fall to 1.14 (0.97–1.30). Current levels of incidence were last seen during the growing phase of the pandemic at the very end of September and will, consequently, require careful monitoring.

From the end of March onwards, the incidence of deaths has continued to fall more sharply than predicted by the model, which is now suggesting a gradual rise over the coming few weeks. This implies that the ONS estimates and the data on deaths are giving conflicting signals and, while this lack of fit has been reduced this week, further model development is required to try and address this discrepancy.

The greatest impact of the changed assumptions regarding the efficacy of the vaccines is in the estimated age-specific probabilities of death given infection (infection fatality rate, IFR). Corresponding plots now show a sharper decline in the IFR once the immunisation programme begins to have an impact in late January. From the end of January we estimate a decreasing IFR in all adult age groups, but most steeply in the older ages. This drop measures the benefits of immunisation against death over and above the benefits against infection. Specifically, there is an estimated fall to a still-high 2.1% in the over-75s and 0.08% overall. The overall impact of the immunisation programme can be seen more clearly in the ‘All Ages’ plot, where the precipitous decline in IFR since late January is a product of this efficacy against death but also of the increasing proportion of infections in young people; older age groups are immunised and become protected against infection. The impact of the second immunisation doses in the 25-44 age-group is beginning to become apparent with a fall after an initial plateau.

For context, alongside the data used here, other indicators (e.g. hospital admissions, reported new positive tests) are suggesting a resurgent epidemic, largely due to the increasing dominance and spread of the Delta strain. Prevalence of infection, as estimated by the ONS Coronavirus Infections Survey, is close to 0.20% in England, though there is large regional heterogeneity with a plateau and perhaps even a slight downturn in many regions. Again, as we consider the possible lifting of social-distancing measures with an Rt around 1, there is the potential for the epidemic to display a range of qualitative behaviours over the coming period. The next few weeks will be crucial. We will continue to monitor the situation closely.

Summary

Real-time tracking of an epidemic, as data accumulate over time, is an essential component of a public health response to a new outbreak. A team of statistical modellers at the MRC Biostatistics Unit (BSU), University of Cambridge, are working to provide regular now-casts and forecasts of COVID-19 infections and deaths. This information feeds directly to the SAGE sub-group, Scientific Pandemic Influenza sub-group on Modelling (SPI-M), and to regional Public Health England (PHE) teams.

Methods

We fit a transmission model (Birrell et al. 2020) to a number of data sources (see ‘Data Sources’), to reconstruct the number of new COVID-19 infections over time in different age groups and NHS regions, estimate a measure of ongoing transmission and predict the number of new COVID-19 deaths.

Data sources

We use:

  1. Data on COVID-19 confirmed deaths from the Public Health England (PHE) line-listing This consists of a combination of deaths notified to:
    • the Demographics Batch Service (DBS), a mechanism that allows PHE to submit a file of patient information to the National Health Service spine for tracing against the personal demographics service (PDS). PHE submit a line list of patients diagnosed with COVID-19 to DBS daily. The file is returned with a death flag and date of death updated (started 20th March, 2020).
    • NHS England, who report data from NHS trusts relating to patients who have died after admission to hospital or within emergency department settings.
    • Health Protection Teams (HPTs), resulting from a select survey created by PHE to capture deaths occurring outside of hospital settings, e.g. care homes (started 23rd March, 2020)
  2. Data on antibody prevalence in blood samples from a PHE survey of NHS Blood Transfusion (NHSBT) donors.

Data are stratified into eight age groups: <1, 1-4, 5-14, 15-24, 25-44, 45-64, 65-74, 75+, and the NHS England regions (North East and Yorkshire, North West, Midlands, East of England, London, South East, South West).

  1. Published information on the the natural history of COVID-19 (Verity et al., 2020; Li et al, 2020)
  2. Information on contacts between different age groups from:
    • A Survey that describes relative rates of contacts between different age groups (Mossong et al. 2008).
    • Google Community Mobility reports, informing the changes in people’s mobility over the course of the pandemic, particularly after the March 23rd lockdown measures.
    • The ONS’ time use survey, which in conjunction with the google mobility study, allows estimation of the changing exposure to infection risk over time.
    • Data from the Department for Education describing the proportion of children currently attending school.
  3. Daily data on the numbers of people getting immunised by age-group and region. These data are derived from the National Immunisation Management Service (NIMS). These data includes all COVID-19 immunisations administered at hospital hubs, local immunisation service sites such as GP practices, and dedicated immunisation centres.

Epidemic summary

Current \(R_t\)

Value of \(R_t\), the average number of secondary infections due to a typical infection today.

Number of infections

Attack rate

The percentage of a given group that has been infected.

By region

By age

Current IFR

Change in infections incidence

Growth rates

NB: negative growth rates are rates of decline. Values are daily changes.

Region Median 95% CrI (lower) 95% CrI (upper)
England 0.01 0.00 0.02
East of England 0.00 -0.03 0.02
East Midlands -0.01 -0.03 0.02
London 0.00 -0.03 0.02
North East 0.00 -0.02 0.03
North West 0.02 0.00 0.04
South East 0.01 -0.02 0.03
South West 0.01 -0.02 0.03
West Midlands 0.00 -0.02 0.03
Yorkshire and The Humber 0.00 -0.03 0.02

Halving times

Halving times in days, if a region shows growth than value will be NA.

Region Median 95% CrI (lower) 95% CrI (upper)
England NA 964.77 NA
East of England 214.24 22.24 NA
East Midlands 100.22 21.61 NA
London NA 27.23 NA
North East NA 30.27 NA
North West NA 153.31 NA
South East NA 33.85 NA
South West NA 38.45 NA
West Midlands NA 28.30 NA
Yorkshire and The Humber 363.94 25.53 NA

Doubling times

Doubling times in days, if a region shows decline then the value will be NA.

Region Median 95% CrI (lower) 95% CrI (upper)
England 78.89 39.72 NA
East of England NA 38.60 NA
East Midlands NA 41.08 NA
London 2365.67 31.64 NA
North East 267.22 24.71 NA
North West 36.90 18.35 NA
South East 96.28 21.86 NA
South West 75.42 21.12 NA
West Midlands 443.37 26.12 NA
Yorkshire and The Humber NA 32.94 NA

Change in deaths incidence

Growth rates

NB: negative growth rates are rates of decline. Values are daily changes.

Region Median 95% CrI (lower) 95% CrI (upper)
England 0.02 0.01 0.02
East of England 0.01 -0.02 0.03
East Midlands 0.00 -0.02 0.03
London 0.01 -0.01 0.03
North East 0.01 -0.01 0.04
North West 0.03 0.01 0.05
South East 0.02 -0.01 0.04
South West 0.02 -0.01 0.04
West Midlands 0.01 -0.01 0.03
Yorkshire and The Humber 0.01 -0.01 0.03

Halving times

Halving times in days, if a region shows growth than value will be NA.

Region Median 95% CrI (lower) 95% CrI (upper)
England NA NA NA
East of England NA 44.53 NA
East Midlands NA 41.27 NA
London NA 73.20 NA
North East NA 74.40 NA
North West NA NA NA
South East NA 78.40 NA
South West NA 123.18 NA
West Midlands NA 71.44 NA
Yorkshire and The Humber NA 57.50 NA

Doubling times

Doubling times in days, if a region shows decline then the value will be NA.

Region Median 95% CrI (lower) 95% CrI (upper)
England 41.61 27.80 76.95
East of England 118.32 24.53 NA
East Midlands 170.95 26.29 NA
London 67.91 22.67 NA
North East 54.30 18.61 NA
North West 20.73 12.63 55.34
South East 44.26 17.98 NA
South West 42.11 17.04 NA
West Midlands 64.01 21.05 NA
Yorkshire and The Humber 96.67 24.15 NA

Infections and deaths

The shaded areas show periods of national lockdown, the green lines the dates (once confirmed) of the steps in the roadmap in the UK Governement’s COVID-19 Response – Spring 2021, and the red line shows the date these results were produced (19 Jun).

Infection incidence

By region

By age

Cumulative infections

By region

By age

Deaths incidence

By region

By age

Cumulative deaths

By region

By age

IFR

Prob \(R_t > 1\)

The figure below shows the probability that \(R_t\) is greater than 1 (ie: the number of infections is growing) in each region over time. Clicking the regions in the legend allows lines to be added or removed from the figure.

\(R_t\)

Copyright © MRC Biostatistics Unit, University of Cambridge